TPS-RPM Registration in Application of MS Lesions MRI Brain Images

Correspondence problem and transformation function are the main challenges in registration so we try to solve these problems concurrently to achieve reasonable and optimum results. We employ non-rigid registration as a non-rigid point matching problem. We have also used the TPS–RPM algorithm. In addition the binary correspondence improved to fuzzy correspondence through the meaning of soft-assign and deterministic annealing. TPS-RPM is applied on MS lesion in MRI images which achieved consecutive weeks to follow up the fluctuation. We compare the iterated closest point ICP’s result to TPSRPM in order to demonstrate the power of the TPS-RPM. Our comparison is done by Evaluated metrics such as Dice similarity measure and accuracy obtained 0.87 and 0.88 in ICP, respectively. For TPS-RPM the Dice similarity measure and accuracy are 0.9 and 0.89, respectively.

Keywords:

non-rigid matching , Registration , TPS-RPM , MS kesion